EURASIP Journal on Audio, Speech, and Music Processing

Table 7 F-scores after 40 and 120 training utterances for DOMOTICA-3

From: The self-taught vocal interface

				Speakers															Average
				17	28	29	30	31	33	34	35	37	40	41	43	44	46	48
			GMM 10	60	51	53	41	43	58	52	53	62	43	53	88	45	83	77	57.5
			GMM 80	53	45	47	34	34	65	46	49	56	37	46	87	37	81	79	53.1
			NMF	90	69	83	82	71	87	76	83	84	73	77	99	75	96	98	82.9
	Compo-
	sitional	N=120	GMM 10	75	66	66	51	52	67	68	66	69	57	66	88	75	88	85	69.3
			GMM 80	80	65	76	56	59	82	74	73	79	69	68	97	78	96	93	76.3
			NMF	99	88	90	93	86	93	91	94	94	92	96	100	99	97	99	94.1
DOMOTICA-3
RØDE M2
	Flat	N=40	GMM 10	43	37	43	23	25	41	38	43	48	29	35	78	62	73	66	45.6
			GMM 80	24	19	27	18	12	45	23	31	42	16	28	84	57	75	74	38.3
			NMF	88	72	80	76	63	78	79	81	77	71	68	99	98	96	98	81.6
		N=120	GMM 10	65	55	65	39	39	48	61	57	51	51	48	91	88	82	81	61.4
			GMM 80	65	50	74	49	35	67	59	61	65	53	60	97	97	85	87	66.9
			NMF	98	83	95	94	78	85	90	89	88	86	93	100	100	100	99	91.9
		N=40	GMM 10	58	45	54	42	42	40	45	53	60	35	56	87	59	86	82	56.3
			GMM 80	54	42	49	37	33	51	46	49	58	31	50	87	54	82	82	53.7
	Compo-		NMF	89	80	89	80	69	79	80	85	86	60	88	99	90	96	98	84.5
	sitional
		N=120	GMM 10	74	56	69	48	52	43	63	69	65	46	69	87	74	90	88	66.2
			GMM 80	79	65	79	57	58	70	74	75	79	55	75	97	89	96	92	76
			NMF	98	92	97	92	86	92	94	96	95	87	98	100	99	99	100	95
DOMOTICA-3
headset
	Flat	N=40	GMM 10	41	29	39	24	24	25	36	41	45	25	37	82	70	73	75	44.4
			GMM 80	27	18	32	18	13	31	23	30	38	11	30	85	59	75	74	37.6
			NMF	88	79	90	78	63	66	83	87	79	52	84	98	98	98	99	82.8
		N=120	GMM 10	64	48	67	45	38	32	57	55	48	38	61	80	89	80	80	58.8
			GMM 80	70	45	73	52	39	52	61	58	64	39	63	96	97	82	89	65.3
			NMF	98	88	97	94	82	92	94	92	92	74	98	100	100	100	100	93.4

The F-scores are interpolated using the LOWESS procedure.

Back to article page